Construction of intelligent prediction model of remaining oil distribution in polymer flooding reservoir under complex geological conditions

Aiming at the challenge of remaining oil distribution prediction in polymer flooding reservoirs under complex geological conditions, this paper proposes an intelligent prediction model based on multi-source data fusion and hybrid machine learning architecture. Polymer flooding technology can improve oil recovery by increasing the viscosity of displacement phase, but under complex geological conditions, the complexity and uncertainty of remaining oil distribution increase, and the accuracy of traditional prediction methods is limited. Therefore, this study integrates static geological data, dynamic production data, geophysical data and displacement experimental data, and uses geostatistical interpolation and deep learning feature extraction to fuse the data. By quantifying geological heterogeneity, such as permeability variation coefficient, interlayer shielding index and other characteristic parameters, the adaptability of the model to complex geological conditions is enhanced The framework of hybrid machine learning model includes input layer, feature crossing layer, LSTM time sequence module, physical constraint layer and output layer, and geological feature encoder and engineering parameter encoder are introduced for data processing. The model adopts transfer learning strategy, and enhances the generalization ability through pre-training in similar blocks and fine-tuning in target blocks. Through the verification of actual reservoir cases, the model is superior to traditional methods in prediction accuracy and adaptability, which provides reliable technical support for reservoir development under complex geological conditions.